Regional economic resilience in the European Union: a numerical general equilibrium analysis

Authors: Di Pietro, F., Lecca, P., Salotti, S. Joint Research Centre 2020 Regional economic resilience in the European Union: a numerical general equilibrium analysis JRC Working Papers on Territorial Modelling and Analysis No 03/2020 JRC TECHNICAL REPORT This publication is a Technical report by the Joint Research Centre (JRC), the European Commission’s science and knowledge service. It aims to provide evidence-based scientific support to the European policymaking process. The scientific output expressed does not imply a policy position of the European Commission. Neither the European Commission nor any person acting on behalf of the Commission is responsible for the use that might be made of this publication. For information on the methodology and quality underlying the data used in this publication for which the source is neither Eurostat nor other Commission services, users should contact the referenced source. The designations employed and the presentation of material on the maps do not imply the expression of any opinion whatsoever on the part of the European Union concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. Contact information Name: Simone Salotti Address: Edificio Expo, C/Inca Garcilaso 3, 41092 Sevilla (Spain) Email: [email protected] Tel.: +34 954488250 EU Science Hub https://ec.europa.eu/jrc JRC120508 Seville: European Commission, 2020 © European Union, 2020 The reuse policy of the European Commission is implemented by the Commission Decision 2011/833/EU of 12 December 2011 on the reuse of Commission documents (OJ L 330, 14.12.2011, p. 39). Except otherwise noted, the reuse of this document is authorised under the Creative Commons Attribution 4.0 International (CC BY 4.0) licence (https://creativecommons.org/licenses/by/4.0/). This means that reuse is allowed provided appropriate credit is given and any changes are indicated. For any use or reproduction of photos or other material that is not owned by the EU, permission must be sought directly from the copyright holders. All content © European Union, 2020 (unless otherwise specified) How to cite this report: Di Pietro, F., Lecca, P., and Salotti, S. (2020). Regional economic resilience in the European Union: a numerical general equilibrium analysis. JRC Working Papers on Territorial Modelling and Analysis No. 03/2020, European Commission, Seville, JRC120508. The JRC Working Papers on Territorial Modelling and Analysis are published under the supervision of Simone Salotti and Andrea Conte of JRC Seville, European Commission. This series mainly addresses the economic analysis related to the regional and territorial policies carried out in the European Union. The Working Papers of the series are mainly targeted to policy analysts and to the academic community and are to be considered as early-stage scientific papers containing relevant policy implications. They are meant to communicate to a broad audience preliminary research findings and to generate a debate and attract feedback for further improvements. Abstract Using a spatial general equilibrium model, this paper investigates the resilience of EU regions under three alternative recessionary shocks, each of them activating different economic adjustments and mechanisms. Using a theoretical framework, we measure the vulnerability, resistance, and recoverability of regions and we identify key regional features affecting the ability of regions to withstand to and recover from unexpected external shocks. The analysis reveals that the response of regions varies according to the nature of the external disturbance and to the pre-shock regional characteristics. Finally, it seems that resilience also depends on factors' mobility. 1 Regional economic resilience in the European Union: a numerical general equilibrium analysis Filippo Di Pietroa, Patrizio Leccab, and Simone Salottib Disclaimer: The views expressed are purely those of the authors and may not in any circumstances be regarded as stating an official position of the European Commission. a: Universidad de Sevilla, Seville, Spain. b: European Commission, Joint Research Centre (JRC), Seville, Spain. Keywords: computable general equilibrium model; regional economic resilience; economic shocks. JEL codes: C68; R13; R15. 1. Introduction Regional economic resilience is a term used to broadly describe how regional economies respond to undesired external disturbances. Essentially, the notion of regional resilience emphasises the ability of regions to resist and recover from shocks and has recently gained popularity among both academics (see for example Fingleton et al. 2015) and policy makers (Šucha et al., 2015, Alessi et al. 2018). Despite the attention drawn to the topic, there is no unique definition of economic resilience in the context of regions (Christopherson. et al., 2010) and the existing studies vary wildly in terms of methods, units of analysis, and aspects of resilience analysed. This paper explores the regional economic resilience of the NUTS-2 (European Nomenclature of Territorial Units for Statistics at level two) regions of the European Union (EU) using the spatial numerical general equilibrium model RHOMOLO (Lecca et al., 2018). It implements a variety of experiments with the purpose of investigating and quantifying the degree of resistance and recoverability of EU regions after an external perturbation. An economic model is appealing to study economic resilience because it overcomes the limitation of case studies and empirical analyses. Only when a shock arises and is perfectly identified, it is possible to determine whether the evolution of regional adjustments is consistent with a resilience path (Sensier et al., 2016). The simultaneous variety of shocks affecting an economy can be challenging to disentangle and additional shocks and disturbances at a later stage can blur the recovery path of the economies hit by the crisis in the first place. A general equilibrium framework permits to reduce this type of complexity and simplify significantly the analysis. In a conventional general equilibrium modelling framework, resilience is seen as the economic system's ability to recover from an external disturbance. The speed at which the economy adjusts to the pre-shock steady-state equilibrium is one of the main objects of interest in this context. Various alternative external shocks (for instance, supply-side or demand-side shocks) can be implemented to test whether regions respond differently 2 to disruptions that are intrinsically different in nature. Essentially, regions are expected to respond differently in relation to the nature of the shock. 1 In our analysis, we simulate three different scenarios, each one with a different type of negative shock contemporaneously hitting all the European NUTS2 regions featured in the RHOMOLO model. The first involves a temporary fall in TFP; the second implies a temporary reduction in the demand for exports to the Rest of the World (ROW); and the third shock consists of an increase of the user cost of capital through a temporary increase in the risk premium. The distinctive feature of this experiment is that in each case we analyse the response of the economy under alternative external disturbances triggering different economic mechanisms. A TFP shock immediately changes the economic structure of regions by directly affecting the supply-side of the economy; a change in exports to ROW implies direct demand-side effects; and a change in the risk premium entails a combination of demand- and supply-side effects. We then use the data coming out of the model simulations in order to study regional economic resilience by constructing variables capturing how the regional economies respond to the shocks and how they behave when recovering from them. Finally, we investigate the main regional characteristics influencing the ability of regional economies to resist and recover after an unexpected external shock. The remainder of the paper is as follows. Section 2 discusses the multi-faceted concept of regional economic resilience. Section 3 briefly presents the RHOMOLO model and section 4 illustrates the strategy adopted for the regional resilience analysis. Section 5 presents and discusses the modelling results in terms of resistance and recoverability of regions. In section 6 we attempt to identify some of the key determinants of resilience, while section 7 explores the potential role played by factor mobility. Finally, section 8 concludes. 2. Regional economic resilience As noted above, there is no unique and commonly accepted definition of regional economic resilience; rather, this highly complex concept has been expanded and analysed along a number of dimensions. Martin (2012) identified three of them: resistance (sensitivity to economic shock impacts), recoverability (the extent and nature of recovery), and reorientation/renewal (the ability of a region to adapt in response to a shock and return to its long-run growth path). Later works (Martin and Sunley, 2015; Martin et al., 2016; Giannakis and Bruggeman, 2017) postulated the existence of additional dimensions such as vulnerability to describe the sensitivity to different types of shocks, and robustness, that is how firms, workers, and institutions respond and adapt to shocks. The existing literature offers several contributions exploring one or more of those aspects of regional economic resilience, mostly using case studies and concentrating on two main dimensions: sensitivity to the shock and recoverability from it (for example, Rizzi et al., 2018, define these two dimensions as the shock and recovery phases). Faggian et al. (2018) analyse the resistance and recovery of the Italian local labour systems during and after the Great Recession, claiming that renewal could only be studied with firm-level data (the use of individual-level data offers a number of additional avenues for research 1 See for example Faggian et al. (2018, p. 396): “A region which might be resilient to a certain type of shock might not be to another type”. 3 as demonstrated by Doran and Fingleton 2015 and 2016). Fingleton et al. (2012) and Martin et al. (2016) study the same two dimensions, resistance and recoverability, in the UK regions during and after the four major recessions of the last four decades concentrating on employment and on the role played by the industrial structure. Crescenzi et al. (2016) analyse the determinants of economic resilience of European regions investigating both national and regional factors. Giannakis and Bruggeman (2017) focus on Greek regions and on the differences between rural and urban ones, while Pudelko et al. (2018) study resistance and recovery after the Great Recession in the regions of Western Germany concentrating on the role played by industrial specialisation. Although this list does not pretend to be exhaustive, it appears that the empirical literature on regional economic resilience is dominated by case studies concentrating on specific regions, shocks, and aspects of resilience (mainly resistance and recovery), while our analysis is based on a different method, that of economic modelling. We offer a novel contribution to the literature by using a general equilibrium model in order to study the economic resilience of the EU NUTS-2 regions. Modelling has already been used to study resilience, but in other contexts mostly pertaining to ecology and disaster studies. For instance, Rose and Liao (2005) study regional resilience in the case of disruption of water services in the Portland Metro economy in Oregon (USA) using a general equilibrium modelling framework. Since we adopt a conventional general equilibrium approach, the simulation experiments presented in the paper are grounded on the so called "engineering resilience" approach (Hill et al., 2008; Martin and Sunley, 2015; and Pudelko et al., 2018) largely inspired by the work done in physical sciences and engineering, rather than evolutionary resilience (Boschma, 2015). In a general equilibrium model, the system is bound to get back to its original equilibrium after a temporary shock, therefore providing the perfect framework to study recovery defined as the return to the pre-shock state, rather than a renewal process modifying the economic structure and relationships within the regional economic systems. The latter concept can be referred to as evolutionary resilience (a concept taken from the ecology field) and it involves structural and operational adaptation in response to shocks, with economies bouncing forward rather than bouncing back (Martin and Sunley, 2015). This is something that a modelling framework such as ours is not equipped to study and it is related to the renewal concept introduced above. Despite its advantages, our approach has some shortcomings. As just stated, regional economies are constrained to adjust towards a stable equilibrium in the long-run, preventing therefore an investigation of the adaptive capacity of regions to move off their equilibrium growth path. Simmie and Martin (2010) and Martin and Sunley (2015) discuss the problems associated with the "equilibrist" interpretation of resilience pointing out that regional economic resilience should be analysed by adopting an evolutionary perspective that intrinsically emphasise the role of the 'multiple phases' of the adaptive cycle of the regional economic evolution. We do believe however that an "equilibrist" perspective should complement alternative methods based on ecological or evolutionary frameworks in order to improve our understanding of resilience under different angles. We firstly study how the EU regions react to various types of shocks, looking both at their vulnerability and at their resistance by concentrating on the magnitude of the initial short-run impacts of the shocks. In the second part of the analysis, we study the recoverability of regions by looking at the performances of the regional economies in 4 terms of getting back to the original steady-state equilibrium. As explained above, it is common for economic studies to concentrate on these dimensions of regional economic resilience, and the general equilibrium modelling approach is ideal to do so. Finally, we investigate the main regional characteristics that influence the ability of regional economies to resist and recover after an unexpected external shock. In order to do so, we resort to an econometric analysis of the modelling data building on the existing literature to identify a few key variables as drivers of the resilience outcomes observed in the simulations. 3. A condensed description of the RHOMOLO model In this section we outline the main equations governing the model to help the reader to identify the main drivers and determinants of the spatial outcomes generated by the model. This presentation is useful to understand the shocks featured in the scenario analysis explained in section 4, and more details on the RHOMOLO model can be found in Lecca et al. (2018). 2 The model represents a decentralised market economy based on the assumption that producers maximize their profits and consumers maximize the utility derived from their consumption, with market prices adjusting endogenously so to keep supply and demand balanced in all markets. The domestic economy consists of 267 endogenous regions, those forming the EU Member States. The ROW is an exogenous external sector. The model features ten NACE rev.2 economic sectors (agriculture, forestry, and fishing; energy; manufacturing; construction; trade and transport; information and communication; financial activities; R&D; public administration; other services) in which firms operate under a monopolistic competition framework à la Dixit-Stiglitz (1977). Consumers The aggregate consumption level 𝐶𝑟 is directly related to the disposable income 𝑌𝐶𝑟: 𝐶𝑟=(1−𝑠𝑟)𝑌𝐶𝑟 𝑃𝑟𝑐 (1) Where 𝑃𝑟𝑐 is the consumer price index and 𝑠𝑟is the rate of savings. Households consume all varieties of final goods available in the economy. In order to represent love for variety, 𝐶𝑟 is assumed to take the form of a CES function defined as: 𝐶𝑟=(∑∑𝜗𝑟,𝑗,𝑖 (𝑐𝑟,𝑗,𝑖)𝜌𝑐 𝑁𝑟,𝑠 𝑖=1 𝐽 𝑗=1 )1 𝜌𝑐 (2) where 𝑐𝑟,𝑗,𝑖 is the consumption of varieties i=1…N of sector j, in region r, whilst 𝜗𝑟,𝑗,𝑖 is a share of expenditure parameter and 𝜌𝑐=𝜎𝑐−1 𝜎𝑐 , where 𝜎𝑐is the elasticity of substitution. Government 2 Additional documentation can be found at https://ec.europa.eu/jrc/en/rhomolo. The RHOMOLO model has also been used in Lecca et al. (2020), Christensen et al. (2019), Kancs and Lecca (2018), and Di Comite et al. (2018). 5 Government expenditure comprises current spending on goods and services 𝐺𝑟,𝑗 and net transfers to households and firms. Its revenues are generated by labour and capital income taxes, and indirect taxes on production. When a balanced budget is applied, either government consumption or the income tax rates are endogenous. In our default configuration we assume fixed government consumption and no change in tax rates. Firms At the firm level (i.e., for each variety), the production technology is represented by a multilevel CES function. In each sector j, and region r, total production 𝑋𝑟,𝑗=𝐶𝐸𝑆[𝑌𝑟,𝑗,𝑉𝑟,𝑗] is a CES combination of the value added 𝑌𝑟,𝑗 and intermediate inputs 𝑉𝑟,𝑗. In turn 𝑌𝑟,𝑗 and 𝑉𝑟,𝑗 are defined as follow in equations (3) and (4) respectively: 𝑌𝑟,𝑗=𝐴𝑦𝑟,𝑗[𝛿𝑟,𝑗 𝑌∙𝐾𝐷𝑟,𝑗 𝜌𝑗𝑦+(1−𝛿𝑟,𝑗 𝑦)∙𝐿𝐷𝑟,𝑗 𝜌𝑗𝑦]1 𝜌𝑗𝑦−𝐹𝐶𝑟,𝑗 (3) 𝑉𝑟,𝑗=(∑𝑏𝑟,𝑠,𝑗𝑣𝑠,𝑗 𝜌𝑣 𝑠)1 𝜌𝑣 (4) In equation (3), 𝑌𝑟,𝑗, is obtained combining capital 𝐾𝐷𝑟,𝑗 and labour 𝐿𝐷𝑟,𝑗 in a CES function, net of fixed costs 𝐹𝐶𝑟,𝑗. Substitution between the two types of primary factors is governed by the parameter 𝜌𝑗𝑦=𝜎𝑦−1 𝜎𝑦 (where 𝜎𝑦 is the elasticity of substitution) and the share parameter 𝛿𝑗𝑌. The scale parameter 𝐴𝑦𝑟,𝑗 represents the conventional hicks neutral technical change parameter in this production function. The input-output relations are shown in equation (4) where the composite demand for intermediate inputs is again a CES combination of 𝑣𝑠,𝑗, that is the purchase of intermediate inputs of each secors j from the supplier sector s. Input substitution between sectors are determined by the elasticity of substitution 𝜌𝑣and the preference parameter related to the share of expenditure 𝑏𝑟,𝑠,𝑗. From cost minimization we obtain the demand for capital and labour in each sector j, represented in equations (5) and (6). 𝐾𝐷𝑟,𝑗=(𝐴𝑦𝑟,𝑗 𝜌𝑗𝑦∙𝛿𝑟,𝑗 𝑦∙𝑃𝑘𝑟 𝑃𝑦𝑟,𝑗)1 1−𝜌𝑗𝑦∙𝑌𝑟,𝑗 (5) 𝐿𝐷𝑟,𝑗=(𝐴𝑦𝑟,𝑗 𝜌𝑗𝑦(1−𝛿𝑟,𝑗 𝑦)∙ 𝑤𝑟 𝑃𝑦𝑟,𝑗)1 1−𝜌𝑗𝑦∙𝑌𝑟,𝑗 (6) where 𝑃𝑦𝑟,𝑗, 𝑃𝑘𝑟,𝑗 and 𝑤𝑟 are respectively the price of value added, the price of capital and the wage rate. Price mark-ups Goods and services can either be sold in the domestic economy or exported to other regions. On the other hand, firms and consumers can purchase inputs within the region or from external markets. We use a single Armington nest that differentiates between domestic and imported goods and do not differentiate between imports from within the country or within the EU. 6 𝑥𝑟,𝑟′,𝑗=𝜂𝑟,𝑟′,𝑖,(𝑃𝑟′,𝑗 𝑃𝑟,𝑟′,𝑗)𝜎𝑗𝑋𝑟′,𝑗 (7) 𝑥𝑟,𝑟′,𝑖,𝑡 is the demand for each goods and services supplied by regions r, to r', 𝜂𝑟,𝑟′,𝑖, is a calibrated expenditure share, 𝑋𝑟,𝑖 is the Armington aggregate of outputs for each firm in region r, while 𝑃𝑟′,𝑗 is defined as a CES price index as over the market price 𝑃𝒓′,𝑟,𝒊,𝒕. 𝑃𝑟′,𝑗,𝑡=(∑𝑁𝑟,𝑗𝜂𝑟,𝑟′,𝑖,𝑃𝑟,𝑟′,𝑗,𝑡 𝑟′ )1−𝜎𝑗 (8) where the price 𝑃𝑟,𝑟′,𝑗 set by a firm of region r (gross of trade cost 𝜏) selling to region r', for a given sector j, is defined as the optimal mark-up (1 𝜀𝑟,𝑟′,𝑗) over the marginal cost 𝑃𝑟,𝑗 ∗, is given as follows: 𝑃𝒓,𝑟′,𝑗=𝜏𝑟,𝑟′,𝑗𝑃𝑟,𝑗 ∗ 1− 1 𝜀𝑟,𝑟′,𝑗 where (9) 𝜀𝑟,𝑟′,𝑗=𝜎𝑟′,𝑗 (10) The marginal cost includes the cost of production factors and the intermediate price index PIN. 𝑃𝑟,𝑗 ∗=𝑎𝑟,𝑗 𝑦𝑃𝑌𝑟,𝑗+𝑎𝑟,𝑗 𝐼𝑛𝑡𝑃𝐼𝑁𝑟,𝑗 (11) 𝑎𝑟,𝑗 𝑦 and 𝑎𝑟,𝑗 𝐼𝑛𝑡 are the share parameters attached to the value added and intermediate inputs respectively. The configuration of RHOMOLO adopted in this paper uses a Dixit-Stiglitz formulation of the mark-up of firm-level product differentiation with elasticities of substitution equal for all firms and products in the model. The elasticity of substitution 𝜎 is the same in each node of the CES function (between home –and imported), therefore any possible combination between domestic and imported inputs will collapse to a single nest. Furthermore the mark-up does not dependent from the market shares, therefore a single region sell products to all the other regions at the same fob (first-on-board) price, even if consumers in the importing regions can observe different cif (cost, insurance and freight) prices, including iceberg transport costs. Wage setting The RHOMOLO model incorporates imperfect competition in the labour market. We assume a flexible framework that allows one to switch from a wage curve to a Philips curve. Further parameterization also permits to use a dynamic or a static form of wage setting. The general formulation is expressed in log as in equation (12): 𝑟𝑤𝑡=𝑎+𝛼 𝑟𝑤𝑡−1− 𝛽 𝑢𝑡+𝛾∆𝑝𝑡−𝜆(𝑟𝑤𝑡−1−Τ𝑡)−𝜃∆𝑢𝑡 (12) 13 𝑅𝑒𝑐𝑜𝑣𝑒𝑟𝑦𝑟=𝐺𝐷𝑃𝑔𝑟𝑜𝑤𝑡ℎ𝑟Recovery−𝐺𝐷𝑃𝑔𝑟𝑜𝑤𝑡ℎ𝐸𝑈Recovery |𝐺𝐷𝑃𝑔𝑟𝑜𝑤𝑡ℎ𝐸𝑈Recovery| (24) The two measures are constructed in a similar way: they both measure the gap between the average growth in the region and the average growth for the EU as whole. However, they diverge with respect to the time period considered: the contraction and the recovery period. For each region, the contraction period is defined as the time frame from the beginning of the shock (period 1) to the period in which the region reaches its negative GDP peak. The recovery period is assumed to start one period after the peak and to end after six periods. The two measures in equation (23) and (24) are centred around zero. A positive value of Resistance indicates that a region is less affected by the contraction relative to the EU average. Similarly, a positive value of the Recovery shows that the regions have a high recoverability relative to the EU average. For each shock implemented in the model, the Resistance and Recovery measures are calculated and plotted in Figures 4, 5, and 6. Each of them results in the EU regions being positioned in one of four quadrants distinguishing between high resistance-weak recoverability (Topleft), high resistance-high recoverability (top-right), weak resistance-high recoverability (bottom-right) and weak resistance-weak recoverability (bottom-left). A casual inspection of the plots suggests that only in the cases of the TFP and the risk premium shocks we are able to identify a negative correlation between Resistance and Recovery, albeit small. It is difficult to detect some sort of direct linear relationship in the third scenario. This suggests that recessionary shocks directly affecting the supply-side of the economy will influence the resistance and recoverability of regions in the opposite way: regions able to recover faster (slowly) are expected to be less (more) resistant to external perturbations. Furthermore, the plots suggest that the position of regions across the four quadrants is not always the same and change according to the type of shock implemented in the model. This result is in line with Faggian et al. (2018) according to which a region shows different resilience degrees depending on the type of shock. It is however possible to identify a small group of regions that constantly maintain their position in one of the quadrants. In particular, 42 regions are likely to experience higher resistance relative to the EU average and 87 regions recover faster than the EU average regardless the type of shock. Interestingly, a group of 28 regions, 22 of which belong to the UK, show high resistance and recoverability regardless the simulation performed and therefore always appears in the top-right quadrant. On the other hand, only three regions retain their position in the left-bottom quadrant across the thre shocks: namely Yugoiztochen (BG34), Languedoc-Roussillon (FR81), and Croatia (HR). 14 Figure 4. Recovery and Resistance - demand shock Figure 5. Recovery and Resistance - TFP shock AT12 AT13 AT21 AT22 AT31AT32 AT33 AT34 BE21 BE22 BE24 BE25 BE31 BE32BE33 BE35 BG31 BG34 CY00 CZ01 CZ02 CZ03 CZ04 CZ05 CZ06 CZ07 CZ08 DE11 DE12 DE13 DE21 DE24 DE26 DE30 DE40 DE50 DE60 DE71 DE80 DE92 DEA2 DEA3 DEA4 DEA5 DEB3 DEC0 DED2 DED4 DED5 DEE0 DEF0 DEG0 DK01 DK02 DK03 DK04 DK05 EL11 EL12 EL13 EL14 EL21 EL22 EL23 EL24 EL25 EL30 EL41 EL42 EL43 ES11 ES12 ES13 ES21 ES22 ES23 ES24 ES30 ES41 ES42 ES43 ES51 ES52 ES53 ES61 ES62 ES63 ES64 ES70 FI1C FI1D FR10 FR21 FR22 FR23 FR24 FR25 FR26 FR30 FR41 FR42 FR43 FR51 FR52 FR53 FR61 FR62 FR71 FR72 FR81 FR82 HRV HU10 HU21 HU22 HU23 HU31 HU32 HU33 IE01 ITC4 ITF6 ITG2 ITH3 ITH4 ITH5 ITI1 ITI4 LU00 LV00 MT00 NL11 NL12 NL13 NL21NL22 NL23 NL31 NL32 NL33 NL34 NL41 NL42 PL11 PL21 PL22 PL31 PL32 PL33 PL34 PL41 PL42 PL43 PL51 PL52 PL61 PL62 PL63 PT11 PT15 PT16 PT17 PT18 PT20 PT30 RO11 RO12 RO21 RO22 RO41 RO42 SE11 SE12 SE21 SE22 SE23 SE31 SE32 SE33 SI01 SI02 SK01 SK02 SK03 SK04 UKD1 UKD3 UKD4 UKE2 UKE4 UKF2 UKG1 UKG2 UKH2 UKH3 UKJ4 UKL2 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 -1.5 -1 -0.5 0 0.5 1 1.5 High Resistance - Weak High Resistance - High recoverability Weak Resistance - Weak recoverability Weak Resistance - High recoverability AT11 AT12 AT13 AT21 AT22 AT31 AT32 AT33 AT34 BE10 BE21 BE22 BE23 BE24 BE25 BE31 BE32 BE33 BE34BE35 BG31 BG32 BG33 BG34 BG41 BG42 CY00 CZ01 CZ02 CZ03 CZ04 CZ05 CZ06 CZ07 CZ08 DE11 DE12 DE13 DE14 DE21 DE22 DE23 DE24 DE25 DE26 DE27 DE30 DE40 DE50 DE60 DE71 DE72 DE73 DE80 DE91 DE92 DE93 DE94 DEA1 DEA2 DEA3 DEA4 DEA5 DEB1 DEB2 DEB3 DEC0 DED2 DED4 DED5 DEE0 DEF0 DEG0 DK01 DK02 DK03 DK04 DK05 EE00 EL11 EL12 EL13 EL14 EL21 EL22 EL23 EL24 EL25 EL30 EL41 EL42 EL43 ES11 ES12 ES13 ES21 ES22 ES23 ES24 ES30 ES41 ES42 ES43 ES51 ES52 ES53 ES61 ES62 ES63 ES64 ES70 FI19 FI20 FI1B FI1C FI1D FR10 FR21 FR22 FR23 FR24FR25FR26 FR30 FR41 FR42 FR43 FR51 FR52 FR53 FR61 FR62 FR63 FR71 FR72 FR81 FR82 FR83 HRV HU10 HU21 HU22 HU23 HU31 HU32 HU33 IE01 IE02 ITC1 ITC3 ITC4 ITF4 ITF6 ITH1 ITH2 ITH3 ITH4 ITH5 ITI1 ITI2 ITI3 ITI4 LT00 LU00 LV00 MT00 NL11 NL12 NL13 NL21 NL22 NL23 NL31 NL32 NL33 NL34 NL41 NL42 PL11 PL12 PL21 PL22 PL31 PL32 PL33 PL34 PL41 PL42 PL43 PL51PL52 PL61 PL62 PL63 PT11 PT15 PT16 PT17 PT18 PT20 PT30 RO11 RO12 RO21 RO22 RO31 RO32 RO41 RO42 SE11 SE12 SE21 SE22 SE23 SE31 SE32 SE33 SI01 SI02 SK01 SK02 SK03 SK04 UKC1 UKC2 UKD1 UKD3 UKD4 UKD6 UKD7 UKE1 UKE2 UKE3 UKE4 UKF1 UKF2 UKF3 UKG1 UKG2UKG3 UKH1 UKH2 UKH3 UKI1 UKI2 UKJ1 UKJ2 UKJ3 UKJ4 UKK1 UKK2 UKK3 UKK4 UKL1 UKL2 UKM2 UKM3 UKM6 UKN0 -0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5 -0.5 -0.3 -0.1 0.1 0.3 0.5 High Resistance - Weak High Resistance - High recoverability Weak Resistance - Weak recoverability Weak Resistance - High recoverability 15 Figure 6. Recovery and Resistance - risk premium shock Although informative, Figures 4-6 do not shed light on the reasons behind the different economic performances of the EU regions following a negative shock. The next section uses two different econometric models built on the model's simulated data (for which Figures 4-6 constitute an early visual exploration) to study the main drivers and determinants of resistance and recoverability. 6. The determinants of resilience In this section, we investigate to what extent regional characteristics influence the ability of regional economies to resist and recover after an unexpected external shock. This is one of the main questions that the scientific literature is trying to answer, and it is also of major importance from a policy-making point of view. Are the most resilient regions more or less open to trade? Or, as predicted by some existing studies, the more specialized the regional economic structure, the lower the resistance of regions? These are key questions to understand the phenomenon of regional economic resilience. While the main aspects of the economic adjustments after each type of shock are by and large common to all regions, the responses to the shocks can differ across regions both in terms of the quantitative impact during the perturbation period (resistance) and in terms of time required to get back to the equilibrium (recoverability). This is endogenously determined in the model and it is affected by the regional initial conditions and the calibrated base year steady-state. We resort to two econometric models based on the model's simulated data presented above in order to identify the main drivers of resistance and recoverability. We do this for two main reasons. First, given that we are working with 267 regions with a significant number of endogenous variables, a clear and comprehensible presentation of the results AT11 AT12 AT13AT21 AT22 AT31 AT32 AT33 AT34 BE10 BE21 BE22 BE23 BE24 BE31 BE32 BE33 BE34 BE35 BG31 BG34 BG41 BG42 CY00 CZ01 CZ02 CZ03 CZ04 CZ05 CZ06 CZ07 CZ08 DE11 DE12 DE13 DE14 DE21 DE23 DE25 DE26 DE27 DE30 DE50 DE60 DE71 DE80 DE92 DE93 DEA1 DEA2 DEA3 DEA4 DEA5 DEB2 DEB3 DEC0 DED2 DED4 DED5 DEE0 DEF0 DEG0 DK01 DK02 DK03 DK04 DK05 ES21 ES22 FI19 FI20 FI1B FI1C FI1D FR10 FR21 FR22 FR23 FR24 FR25 FR26 FR30 FR41 FR42 FR43 FR51 FR52 FR53 FR61 FR62 FR63 FR71 FR72 FR81 FR82 FR83 HRV HU10 HU21 HU22 HU23 HU31 HU32 HU33 IE01 ITC4 ITF4 ITF6 ITG1 ITH1 ITH2 ITH3 ITH4 ITH5 ITI1 ITI4 LT00 LU00 LV00 MT00 NL11 NL12 NL13 NL21 NL22 NL23 NL31 NL32 NL33 NL34 NL41 NL42 PL11 PL21 PL31 PL32 PL34 PL42 PL43 PL51 PL62 PL63 PT11 PT15 PT16 PT17 PT18 PT20 RO11 RO12 RO21 RO22 RO31 RO32 RO41 RO42 SE11 SE12 SE21 SE22 SE23 SE31 SE32 SE33 SI01 SI02 SK01 UKD1 UKD3 UKD4 UKE2 UKE3 UKE4 UKF1 UKF2 UKG1 UKG2 UKG3 UKH1 UKH2 UKH3 UKI1 UKI2 UKJ2 UKJ4 UKK2 UKL1 UKL2 UKM3 UKM6 UKN0 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 -1.5 -1 -0.5 0 0.5 1 1.5 High Resistance - Weak High Resistance - High recoverability Weak Resistance - Weak recoverability Weak Resistance - High recoverability 16 is problematic. The high dimensionality of the model prevents a more straightforward presentational approach, therefore it is easier to perform appropriate regression analyses on the model's simulated data to gain insights of the average behaviour of regional economies under alternative initial characteristics. Second, the econometric analysis helps to generalize and better summarize the results. Thus, we do not focus on the results of specific regions, but rather on the average effects driven by changes in the initial regional characteristics. We begin by analysing the regional characteristics that affect the resistance of regions, and subsequently we turn our attention to their recoverability. Both analyses are done separately for the three different kinds of shocks, allowing for the role of each determinant to depend on the nature of the initial perturbation. A set of economic resilience potential determinants have been selected through an initial screening of bivariate correlations involving variables already considered in previous works on economic resilience. This helped us to exclude redundant variables, focusing more on the most promising drivers of resilience. These are listed below. Factor intensity. It indicates whether the regional production process is more or less capital (or labour) intensive. It is reasonable to expect that regions relatively more capital intensive could experience a bigger drop in economic activities when the recessionary shock directly affects investments and capital adjustments. Variables measuring either capital or labour intensity have been used for similar analyses by, among others, Briguglio et al. (2009) and Rizzi et al. (2017). Openness. The degree to which an economy depends on foreign trade affect the vulnerability of regions (Briguglio, 1995 and Briguglio et al., 2009). Therefore, regions that are more open can potentially be less resistant to external shocks because of its dependence to external environments. On the other hand, openness might also be a source of strength thanks to the positive role played by international trade. Specialization. Industrial specialization (diversification) has been widely studied in the context of economic growth (Glaeser et al 1992; Henderson et al., 1995; Harrison et al., 1996; Henderson 2003) and a number of authors have analysed its role as determinant of economic resilience (Frenken et al., 2007, Hill et al., 2008; Fingleton et al., 2012, Pudelko et al., 2018; Faggian et al., 2018; Rizzi et al., 2017 and Lezzetti et al., 2019). Theoretically, highly specialized regions can suffer higher damages when the economic shocks involves the sectors in which they are specialized. It is however unclear what to expect in terms of influence of this variable on resistance and recoverability. For instance, Martin et al. (2016) find that a diverse economic structure may confer greater regional resilience to shocks than more specialized ones. Our empirical strategy starts with the following model for resistance: ∆𝐺𝐷𝑃𝑟𝑝𝑒𝑎𝑘=𝛽0+𝛽1𝐿𝑆𝑟+𝛽2𝑂𝑝𝑒𝑛𝑒𝑠𝑠𝑟+𝛽3𝑆𝐼𝑟+𝛽5𝐶𝐷+𝜀𝑟 (25) where r is the region and єr is the heteroscedastic random error. The dependent variable ∆𝐺𝐷𝑃𝑟𝑝𝑒𝑎𝑘is the change in GDP observed at the time of the negative peak, which is the highest negative change in GDP observed during the period of the shock. Typically, regions experiencing largest negative changes in GDP are showing less resistance to the shock. Although the magnitude and duration of the shock is the same for all regions, the time at which they reach the largest negative impact might diverge, as demonstrated in 17 Table 1. For example, under the TFP shock, the dependent variable always takes the values of the GDP changes recorded one period after the shock (see Table 1). For the other two shocks, the period at which the largest negative changes in GDP is observed varies widely. As for the right-hand side variables, 𝐿𝑆𝑟 is the initial labour share; 𝑂𝑝𝑒𝑛𝑒𝑠𝑠𝑟 is equal to the sum of import plus export divided by the gross domestic product; and 𝑆𝐼𝑟 is the standard Krugman (1991) specialization index adapted to the EU regional context. 5 𝐶𝐷 is a set of country dummies, one of the two alternative ways to control for geographic factors, the other being a regional contiguity variable accounting for the number of shared borders of the 267 regions of the EU (𝐶𝑇𝐺). Table 2 contains some descriptive statistics of the variables used for the estimation of equation (25). The empirical model is estimated using Ordinary Least Square (OLS) with robust standard errors (the results of the Moran’s I test of spatial autocorrelation, not reported but available upon request, do not suggest any spatial dependence issue). 6 Table 2 Summary statistics of the variables used in the econometric model Variables N Mean SD Min Max ∆𝑮𝑫𝑷 TFP 267 -1.01 0.12 -1.28 -0.42 Demand shock 267 -0.07 0.26 -1.00 0.00 Risk-premium shock 267 -0.48 0.35 -1.46 0.00 LS 267 0.58 0.10 0.25 0.74 Openness 267 5.60 1.72 3.55 20.56 SI 267 0.40 0.20 0.09 1.12 Table 3 shows the estimated coefficients associated with the three alternative versions of model (25) estimated separately for the three scenarios of the analysis (TFP shock, demand shock, and risk premium shock). Columns (1) contain the results of the baseline model where the set of explicative variables is constituted only by the labour share 𝐿𝑆, trade openness 𝑂𝑝𝑒𝑛𝑛𝑒𝑠𝑠, and the specialization index 𝑆𝐼. The estimates reported under columns (2) and (3) are obtained by adding to the model either the country dummies or the regional contiguity variable, respectively. A first look at the table shows that most of the coefficients are statistically significant and have the expected signs. Trade openness appears to be negatively related to GDP when either a TFP or a demand shock hits the economy. This means that the higher the initial ratio of exports and imports over GDP, the bigger the loss of GDP caused by the shock. The explanation for this could lie in the more export-oriented regions being more sensitive to these two shocks due to the competitiveness effects of both the change in TFP (through the changes in commodity prices) and of the imposed reduction in the exports to the ROW. On the other hand, the role of openness is not clear in the risk premium scenario as the 5 SI is calculated with the following formula: 𝑆𝑝𝑒𝑐𝑖𝑎𝑙𝑖𝑧𝑎𝑡𝑖𝑜𝑛 𝐼𝑛𝑑𝑒𝑥𝑟=∑| 𝐸𝑖,𝑟 ∑𝐸𝑖,𝑟𝑖 −𝐸𝑖,𝐸𝑈 ∑𝐸𝑖,𝐸𝑈𝑖| 𝑖 6 We did not detect multicollinearity in the regression analysis. The maximum value of the variance inflation factor is lower than 2 in all cases. 18 estimates of its coefficient depend on the inclusion or exclusion of the country dummies and the contiguity variable. Table 3: Regressions results of the resistance model TFP shock Demand shock Risk Premium shock Variables (1) (2) (3) (1) (2) (3) (1) (2) (3) LS -0.111 0.186** -0.117* 0.113*** 0.199*** 0.112*** 2.533*** 3.280*** 2.495*** (0.058) (0.085) (0.058) (0.037) (0.065) (0.008) (0.162) (0.158) (0.159) Openness -0.013*** -0.009*** -0.013*** -0.048*** -0.048*** -0.051*** -0.007 0.016** 0.003 (0.030) (0.003) (0.003) (0.008) (0.009) (0.008) (0.009) (0.009) (-0.009) SI 0.442*** 0.317*** 0.460*** -0.071* -0.081* -0.052 -0.038 0.013 0.055 (0.028) (0.026) (0.029) (0.0358) (0.039) (0.037) (0.077) (0.069) (0.079) constant -1.049*** -1.185*** -1.072*** 0.126 -0.062 0.079 -1.982*** -2.282*** -2.096*** (0.045) (0.067) (0.046) (0.067) (0.119) (0.069) (0.127) (0.126) (0.127) CTG -0.006* 0.014 0.029*** (0.003) (0.007) (0.007) Country dummies no yes no no yes no no yes no No. obs. 267 267 267 267 267 267 267 267 267 R-square 0.49 0.82 0.50 0.22 0.51 0.23 0.52 0.89 0.54 Notes: Standard errors in parentheses; *** indicates significance at the <0.001 level; **<0.01; *<0.05 The coefficients associated with the labour share are positive and significant under the demand and risk premium shock for each of the alternative models estimated. For the TFP shock, the labour share coefficient is positive and significant only when country dummies are included. As for the demand and risk premium scenarios, the estimates imply that the sign of the relationship between the capital share and GDP changes is negative during the perturbation period. This effect is related to the structure of the initial steady-state equilibrium, where export-oriented regions are typically more capital intensive and for this reason will be less resistant than labour intensive regions under external demand shocks. For the case of an increase in the risk premium, the reduced expectations of future profits make both investments and the capital stock to fall. We then observe greater disinvestment effects and therefore larger decreases in capital stock in those regions with higher capital-GDP ratios. This implies that capital-intensive regions are likely to suffer relatively more than those regions with lower capital shares in the original equilibrium. The estimated coefficient of the specialization index is significant and positive in all three variants of the model in the case of the TFP scenario, signalling that regions that are more specialized are likely to be less affected when the shock hits the economy. The opposite is true in the case of the demand shock, although the coefficient is significant only when the contiguity variable is excluded from the model. Finally, results for this variable are inconclusive in the risk premium scenario. We now turn our focus on the main factors driving regional recovery by using a probit model based once again on the model's simulated data. The estimated equation is the following: Pr(𝑌𝑟=1|𝑋𝑟)=𝜃(𝑋𝑟𝑇𝛽) (26) Y is a regional dummy variable that takes the value 1 if the number of periods required to get back to the steady-state equilibrium after the period of the peak is below the average number periods needed in the EU as whole. It takes the value of zero otherwise. 19 𝑋𝑇indicates the set of explanatory variables. In addition to the variables used for the estimation of equation (25), this model also includes ∆𝐺𝐷𝑃𝑝𝑒𝑎𝑘 as defined in equation (25). This is added to the right-hand side variables in order to evaluate whether regions experiencing higher distress (larger drops in GDP) in the aftermath of a shock are also struggling to readjust and recover. The estimated coefficients and marginal effects of the probit model of equation (26) are reported in Table 4. Table 4: Regressions results of the recoverability model TFP shock Demand shock Risk premium shock Coeff. Change in prob. Coeff. Change in prob. Coeff. Change in prob. LS 9.72*** (1.35) 3.11*** 10.98*** (2.26) 3.54*** 24.42*** (4.13) 2.79** Openness -0.04 (0.06) -0.01 -0.42* (0.17) -0.14* 0.30 (0.19) 0.03 SI -0.41 (0.74) -0.13 -1.06* (0.52) -0.34* -1.33* (0.59) -0.15* ΔGDPpeak 13.46*** (2.12) 4.30*** 6.58*** (1.07) 5.92*** 4.27*** (0.66) 0.49** Constant -0.611*** (0.231) -2.58 (1.88) -14.87** (2.98) No. obs. 267 267 267 Pearson χ^2 232.38 188.00 3509.62 Pseudo R^2 0.35 0.58 0.23 Notes: Standard errors in parentheses; *** indicates significance at the <0.001 level; **<0.01; *<0.05 In all scenarios, the marginal effect of the GDP change, following the negative shock, is positive and statistically significant. This means that the larger the loss in GDP (in absolute value), the less probable is for the region to recover its steady-state faster than the EU average. The labour share shows positive marginal effects in all cases, thus larger labour shares are associated with a more rapid recovery. The specialization index and trade openness have mainly negative marginal effects, meaning that higher values of these two variables are associated with a lower probability of making a faster recovery after the negative GDP peak. However, the specialization effects are statistically significant only in two cases out of three (demand and risk premium scenarios), and the openness effect is only significant when estimated on the data arising from the risk premium scenario. Thus, it appears that, at least for certain types of shocks, more open and highly specialized regions are less likely to recover faster than the EU average. The above analysis on the determinants of both regional resistance and recoverability reveals that certain initial conditions are of extreme importance for these two aspects of regional economic resilience. The empirical models built on simulated data suggests that the calibrated shares 7 which govern the initial model's equilibrium can contemporaneously affect the level of resistance and the speed of recovery after negative shocks. Interestingly, but not surprisingly, the legacies effects of the shocks are also related to the magnitude of the impact caused by shock itself. 7. The role of factors mobility In this section, we assess the sensitivity of the results in relations to some of the model’s assumptions. 8 All the results presented so far refer to the model with both capital and 7 The base year data is therefore of utmost importance in this analysis. Given that alternative dataset are not currently available, the results of the model have been generalized using regression models in order to obtain average predicted behaviour of regions with specific characteristics. 8 Reporting the sensitivity of the results to alteration of behavioural parameters has undoubtedly a pedagogical benefit. However, it is relatively easier to guess how variables move in relation to changes in behavioural parameters (e.g., elasticity of substitutions) for the three illustrative recessionary scenarios under investigation. Variations in the trade and production elasticities only 20 labour assumed to be mobile across EU regions. The literature on resilience has highlighted a key role of factor mobility: for example, Yair Grinberger and Samuels (2018) study in a theoretical setting how resilience outcomes change depending on the existence of labour market mobility after a natural disaster. The intuition behind such result is simple: when a region is hit by a shock, factors of productions are likely to quickly leave the region and possibly exacerbate the effects of the initial shock. At the same time, as regions adjust to the shock, new factors of production may flow into the region and help both softening the impact of the shock and accelerating the recovery. In order to study the role of factor mobility, we run three alternative sets of simulations in which we turn off either capital mobility, labour mobility, or both at the same time. We then compute, and report for the EU as whole, the percentage differences in EU GDP under no factor mobility against the three alternative specifications in which there is some mobility (either capital or labour) or full mobility (which is our default assumption). Figures 7, 8 and 9 show the evolution of the GDP differences between these model's specifications over thirty periods, for the TFP shock, demand shock, and risk premium shock respectively. In terms of interpretation, if the observed difference is positive it means that factor mobility makes the region relatively better off than in the case in which the factors of production are not allowed to move across regions. If the GDP differences are negative, factor's mobility exacerbates the effect of the negative shock. In the case of the TFP shock, and excluding the firsts two periods, we typically observe that factor mobility makes regions relatively more vulnerable, reducing both their resistance (larger GDP loss after the impact of the shock) and their recoverability (GPD is lower also in the period after it reaches the negative GDP peak). The position of the curves in Figure 7 also suggests that the vulnerability and the adjustment of the economy towards a steady-state equilibrium at least up to period 25 is more affected by labour mobility rather than capital mobility. Things are different in the demand shock scenario. Capital mobility moderates the negative effects of the shock on EU GDP, while labour mobility exerts the opposite effect for the first 12 periods. When both factors are allowed to move across regions, the negative effects of the shock are exacerbated for the first 20 periods, after which the positive effects of labour mobility dominate the overall impact of factor mobility. In the risk premium scenario, labour mobility moderates the negative effects of the shock on GDP, but its effect is dominated by the negative influence of capital mobility for the first 20 periods when both factors are allowed to move freely. Capital mobility makes the EU economy less vulnerable in the first four periods, but less resilient afterwards. alter, more or less proportionally, the magnitude of the results but the relative position of regions remain substantially unchanged (detailed results not reported for the sake of brevity but available upon request). For this reason, we thought would have been more appropriate to assess the robustness and sensitiveness of the results in terms of modelling assumptions rather than model parametrizations. 21 Figure 7. EU GDP evolution under alternative specifications of factors mobility - TFP shock Figure 8. EU GDP evolution under alternative specifications of factors mobility - Demand shock -0.014 -0.012 -0.010 -0.008 -0.006 -0.004 -0.002 0.000 0.002 0.004 1 4 7 10 13 16 19 22 25 28 Full factors mobility Labour Mobility Capital Mobility -0.008 -0.006 -0.004 -0.002 0.000 0.002 0.004 0.006 1 4 7 10 13 16 19 22 25 28 % differences Full factors mobility Labour Mobility Capital Mobility 22 Figure 9. EU GDP evolution under alternative specifications of factors mobility - Risk premium shock All in all, it appears that capital mobility initially moderates the negative GDP effects of a shock, but as the economy adjusts, it exacerbates the negative effects on GDP and hinders the recoverability of the regional economies. On the other hand, the effect of labour mobility depends on the nature of the shock: it is negative in the event of a TFP shock, positive in the event of a risk premium one, and it's negative for resistance but positive for recoverability when a demand shock hits the economy. 8. Conclusions This paper investigates the likely response of EU regions to three alternative external disturbances, each of them triggering different economic mechanisms: a fall in TFP, a reduction in demand of exports to the ROW, and an increase of the rate of return to capital through an increase in the risk premium. We found significant differences across regions in both their resistance to and recoverability from unexpected recessionary shocks. Furthermore, our results suggest that the regional responses to external perturbations change depending on the nature of the shock. Regions highly resilient to a supply-side recessionary shock could be weakly resilient to demand-side shocks. We also look for the regional characteristics driving the resistance and the recoverability of the economies. We found that regions relatively more open are less resistant under either a TFP or a demand shock while capital intensive regions will be less resistant than labour intensive regions under a risk premium shock. Our analysis also suggests that regions experiencing larger recessionary shocks are also struggling to recover. Moreover, it will be unlikely for regions which are highly specialized and relatively more open to experience a faster recovery after a negative shock. We also analysed the role of factors' mobility finding that capital mobility initially moderates the negative GDP effects of a shock, but then exacerbates its impact a later -0.015 -0.01 -0.005 0 0.005 0.01 0.015 0.02 12345678910 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 Full factors mobility Labour Mobility Capital Mobility